Observer variability in RECIST-based tumour burden measurements: a meta-analysis

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Tumour burden measurements using Response Evaluation Criteria in Solid Tumours (RECIST) are subject to observer variability.
• Quantifying this variability is crucial for accurate treatment response assessment.

Purpose of the Study:

• To determine the extent of observer variability in tumour burden measurements using RECIST.
• To evaluate the impact of different measurement methods on variability.

Main Methods:

• A systematic literature search identified studies on observer variability in RECIST measurements via computed tomography.
• Pooled analysis of 95% limit of agreement (LOA) values for relative measurement difference (RMD) using a random-effects model.

Main Results:

• Pooled 95% LOAs for single lesion longest diameter measurements ranged from -22.1% to 25.4% between observers and -17.8% to 16.1% for a single observer.
• For the sum of multiple lesions, pooled 95% LOAs ranged from -19.2% to 19.5% between observers and -9.8% to 13.1% for a single observer.
• Interobserver variability in interval change calculation for single lesions ranged from -31.3% to 30.3%.

Conclusions:

• Interobserver variability in RECIST measurements, particularly for single lesions and interval changes, can exceed the 20% threshold for progression.
• Variability is reduced when measurements are performed by a single observer or when assessing the sum of multiple lesions.